How to calculate honest ROI on an AI project

The AI business case that gets approved is rarely the one with the biggest number. It is the one the finance person believes.

Count the costs vendors leave out

• Licence cost per seat, at the number of seats you will actually need in month twelve — not month one.
• Implementation and integration time, priced at real internal day rates.
• Training, including the productivity dip while people learn.
• Ongoing oversight — someone checks outputs, and that someone costs money.
• The failure rate: assume a share of outputs are unusable and must be redone.

Model the benefit three ways

Run conservative, expected and optimistic scenarios. Conservative should assume half the time saving and two-thirds of the adoption. If the conservative case still clears your hurdle rate, you have a project. If only the optimistic case works, you have a hope.

A worked shape

A ten-person professional services firm automating document drafting might save four hours per fee earner per week. At a conservative 50% realisation and a £45 blended hourly cost, that is roughly £23,000 a year against maybe £6,000 of licence and oversight cost. The maths works — but only because the hours were measured before the project, not estimated after it.

Time to value matters more than total value

A project returning £15,000 in three months beats one returning £40,000 in eighteen, because the first one funds the second and keeps the organisation believing. Sequence your roadmap by time to value, not by size of prize.